Is Copyleaks accurate? I tested it on 50 essays

Copyleaks is one of the AI detectors my university uses alongside Turnitin, so I wanted to know how accurate it really is. I ran 50 essays through it: 20 fully human-written, 15 fully AI-generated, and 15 mixed (human-written with AI assistance).

Results on Human-Written Essays

Out of 20 human-written essays, Copyleaks correctly identified 17 as human (85% accuracy). Three essays were falsely flagged as partially AI-generated, with AI percentages of 12%, 18%, and 23%. Two of the false positives were from ESL students, which matches what I’ve heard about detectors struggling with non-native English.

Results on AI-Generated Essays

Copyleaks correctly identified 13 out of 15 fully AI-generated essays (87% accuracy). The two it missed were essays that had been lightly edited by hand after generation. Interestingly, it was better at catching ChatGPT output than Claude output.

Results on Mixed Content

This is where things get messy. For the 15 mixed essays (human-written with some AI-assisted sections), Copyleaks gave wildly varying results. Some scored 5% AI, others 65% AI, even when the amount of AI assistance was similar. The tool seems to struggle when human and AI writing are interleaved.

Compared to Turnitin

Overall, I’d say Copyleaks is slightly less accurate than Turnitin for AI detection, but it has the advantage of being more affordable and accessible. Its multilingual support is also better than most competitors, which matters for international students.

Should You Trust It?

No AI detector should be treated as definitive proof. Copyleaks is a useful indicator but the 15% false positive rate on human-written content is concerning. If you get flagged, remember you have the right to appeal with evidence of your writing process.